mcp

Purpose

Perspective AI MCP Server is the MCP (Model Context Protocol) server for Perspective AI. Its core purpose is to enable AI assistants to operate the Perspective AI platform directly within MCP-compatible clients (such as Claude Desktop, Claude Code, Cursor, etc.), replacing traditional static forms with adaptive AI conversations, thereby understanding users' real situations, automatically structuring key information, and triggering the next steps.

Main Capabilities

Once connected, the AI assistant can call the following 22 tools, grouped by lifecycle stage:

Workspace Management

• workspace_list: List all accessible workspaces
• workspace_get: Get workspace details
• workspace_get_default: Get the default workspace

Design and Iteration

• perspective_list: List perspectives in a workspace, or search across workspaces by name
• perspective_get: Get the full configuration and statistics of a perspective
• perspective_create: Create a new perspective from a natural language brief
• perspective_respond: Answer follow-up questions during the design process
• perspective_update: Refine a perspective through natural language feedback
• perspective_await_job: Long-poll to wait for a design task to complete
• perspective_get_preview_link: Get a shareable preview link for pre-deployment testing

Deployment and Distribution

• perspective_get_embed_options: Get embed code snippets (fullpage, widget, popup, slider, float, card) and SDK reference
• participant_invite: Create a magic link invitation valid for 48 hours, optionally sent by email

Analytics

• perspective_get_stats: Get aggregated statistics and distributions for a time period
• perspective_list_conversations: Byfilter conditions (status, trust score, date) list conversations
• perspective_get_conversation: Get full conversation details, including transcript and trust assessment
• perspective_get_conversations: Efficiently fetch conversations in bulk for batch analysis

Automation

• automation_list: List the automations of a perspective along with their status, channels, and metadata
• automation_create: Create automations (Webhook, email, Slack, HubSpot)
• automation_update: Update fields of an existing automation
• automation_delete: Permanently delete an automation
• automation_test: Run end-to-end tests against a simulated conversation
• integration_manage: List providers (Slack, HubSpot), connect them, and search their tools

How to Use

Connect via OAuth (Recommended)

Add the MCP URL to your client and complete a one-time browser login. Tokens are not stored in configuration files; each connected app appears in Settings → Connected Apps and can be revoked at any time. The MCP URL is the same for everyone.

• Cursor (marketplace, recommended): Open Customize in the sidebar, search the marketplace for "Perspective AI", and follow the prompts to complete the OAuth flow. The plugin includes the MCP server configuration and a skill that helps the AI assistant determine when to use Perspective tools.
• Cursor and VS Code (manual): Add to mcp.json.
• Claude.ai (web and desktop): Open the Add custom connector dialog and fill it in, then click Add to complete the OAuth browser flow. (Manual path: Customize → Connectors → Add custom connector. Available on Free, Pro, Max, Team, and Enterprise plans.)
• Other MCP clients: Clients that only support stdio can use mcp-remote, which automatically discovers OAuth metadata. On the first call to a Perspective tool, the client will open a browserComplete OAuth in the browser window, and subsequent requests reuse the token.

Connect via Personal Access Token

If the client does not support OAuth, or if you want to use long-lived token authentication, you can generate a token in Perspective AI Settings → MCP.

One-click install for Claude Desktop

Download perspective.mcpb, double-click to open it in Claude Desktop, click Install, and paste the token as prompted.

Build from source

Building will output perspective.mcpb in the current directory.

Use cases

• Design: Create four conversation agent types—Interviewer (scaled in-depth qualitative interviews), Concierge (replace static forms with conversational flows), Evaluator (turn boring surveys into conversations), Advocate (listen first, then respond on behalf of a position, brand, or cause). Browse 180+ templates or get inspiration from use cases by role and industry.
• Deploy: Embed on your website or send personalized invitations to specific participants.
• Analyze: Analyze why conversations are abandoned and summarize feedback.
• Automate: When a conversation's trust score exceeds a threshold, push it to HubSpot and notify Slack.

Example prompts:

• "Design a Concierge that qualifies pricing-page leads by budget and timeline. Keep the tone warm, not salesy."
• "Build an Interviewer for churn research. Find out why customers left, what alternatives they chose, and what would have kept them."
• "Spin up an Evaluator for 30-day onboarding feedback. What's working and what's confusing?"
• "Set up an Advocate for our refund policy. Listen first, then explain the process without sounding defensive."
• "Give me the popup embed snippet for my concierge and walk me through adding it to our pricing page in Next.js."
• "Invite our 20 design partners to my beta-feedback perspective, and prefill each person's name and email."
• "Why are people abandoning my lead-capture concierge this week? Pull the drop-off conversations and summarize the top reasons."
• "Whenever a conversation scores above 80 on trust, push it to HubSpot as a contact and ping #sales in Slack."

Security and data

• OAuth: Tokens issued via OAuth are scoped to the workspace and bound to the connected client, and can be revoked at any time in Settings → Connected Apps.
• Personal Access Token: A PAT is bound to your Perspective AI account, and the authorization server acts on your behalf in your workspace. You can generate, rotate, or revoke it at getperspective.ai/settings/mcp.
• Transport: All traffic is encrypted via TLS. Tokens are sent only via the Authorization header to https://getperspective.ai/mcp。
• Data residency: Conversation data remains within your Perspective AI workspace. The .mcpb package is a lightweight stdio-to-HTTP proxy and does not store or cache responses locally.

Limitations

No information is currently available about limitations beyond billing, pricing, and platform compatibility.